Technical knowledge distance measurement method and equipment based on graph representation learning
Through a graph representation learning method, combining patented metrological data and semantic information, the asymmetric distance between technical knowledge is solved, and the problem that traditional methods are difficult to capture the relationship between complex structures and asymmetric knowledge is achieved, and more accurate knowledge distance measurement is achieved.
Patent Information
- Application Number
- CN202411796193.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-09
- Publication Date
- 2025-04-25
AI Technical Summary
The traditional relationship between patent citations and patent co-occurrence is difficult to accurately capture the knowledge distance in the complex structure of the technical system, especially when the trend of technological fusion is intensified and the citation expansion is expanded.
The technical knowledge distance measurement method based on graph representation learning is adopted, and asymmetric knowledge distance is calculated by obtaining patent bibliographic data, extracting technical knowledge elements and their association relationships, extracting semantic information, and using graph representation learning algorithms to fuse reference vectors, relation vectors and semantic vectors.
This method can more accurately and comprehensively reflect the distance between technical knowledge, solving the problem that traditional methods are difficult to capture the relationship between complex structures and asymmetric knowledge.
Smart Images
Figure CN119886296A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of special data processing, and in particular to a method and device for measuring technical knowledge distance based on graph representation learning. Background Art
[0002] With the emergence of cross-integration phenomena in the scientific and technological innovation system, cross-domain knowledge reorganization has become the main driving force for scientific and technological development. In this context, it is increasingly important to accurately depict the distance between knowledge, which can reflect the degree of separation or difference between reorganized knowledge and reflect the novelty of reorganized innovation. However, in the technology system, traditional knowledge distance measurement methods usually only measure through the citation or co-occurrence relationship of patent data, ignoring the directionality and weight characteristics of knowledge flow, resulting in the inability to accurately depict the complex structure of the technology system. In addition, with the intensification of the trend of technological integration and the phenomenon of citation inflation, the boundaries between technical fields have gradually blurred, and it is difficult to accurately capture the knowledge distance by simply relying on traditional patent citations and patent co-occurrence relationships.
[0003] Therefore, a new measurement method that integrates multiple information is urgently needed to more comprehensively reflect the knowledge distance between technology components. Summary of the invention
[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method and device for measuring technical knowledge distance based on graph representation learning.
[0005] The purpose of the present invention can be achieved by the following technical solutions:
[0006] According to a first aspect of the present invention, a method for measuring technical knowledge distance based on graph representation learning is provided, the method comprising:
[0007] Acquiring patent bibliometric data, and extracting technical knowledge elements based on the patent bibliometric data;
[0008] Acquire the association relationship between the technical knowledge elements, and construct a relationship network based on the association relationship, wherein the association relationship includes co-occurrence and citation, and the relationship network includes a co-occurrence network and a citation network;
[0009] Extract semantic information of each technical knowledge element;
[0010] Based on the association network and the semantic information, a reference vector, a relationship vector and a semantic vector are obtained, and they are fused to obtain a fusion vector;
[0011] The similarity of the fused vectors is calculated, and the asymmetric knowledge distance is calculated based on the similarity.
[0012] As a preferred technical solution, the method for extracting technical knowledge elements is:
[0013] Structuring the patent bibliometric data and obtaining patent classification numbers;
[0014] Identify and extract the main classification number of the patent classification number, and extract technical knowledge elements based on the results of identification and extraction.
[0015] As a preferred technical solution, the method for extracting semantic information of each technical knowledge element is:
[0016] Obtaining semantic interpretations of all sub-classification codes contained in the main classification code;
[0017] The semantic interpretation of each sub-classification number is converted into a character string, and the character strings are concatenated to obtain semantic information.
[0018] As a preferred technical solution, the method for obtaining the fusion vector is:
[0019] Encoding the nodes of the citation network and the co-occurrence network using a graph representation learning algorithm to obtain a co-occurrence vector and a citation vector;
[0020] Extracting a semantic vector based on the semantic information;
[0021] The citation vector, co-occurrence vector and semantic vector are concatenated to obtain a fusion vector.
[0022] As a preferred technical solution, the splicing method is:
[0023] The citation vector, co-occurrence vector and semantic vector are all reduced to the same dimension;
[0024] The citation vectors, co-occurrence vectors and semantic vectors after dimensionality reduction are horizontally spliced by horizontal expansion, wherein the citation vectors, co-occurrence vectors and semantic vectors after dimensionality reduction are all multi-dimensional vectors.
[0025] As a preferred technical solution, the method for calculating the asymmetric knowledge distance is:
[0026] Use cosine similarity to calculate the similarity of the fused vectors;
[0027] Max-Min normalization is used to convert the similarity into an asymmetric knowledge distance, which is expressed as:
[0028]
[0029] Among them, KD i,j KS represents the asymmetric knowledge distance between technical knowledge element j and technical knowledge element i; i,jRepresents the similarity between the fusion vector of knowledge technology element i and the fusion vector of j; Min i Indicates the minimum similarity between the fusion vector of technical knowledge element i and the fusion vectors of other technical knowledge elements; Max i Indicates the maximum similarity between the fusion vector of knowledge element i and the fusion vectors of other technical knowledge elements.
[0030] As a preferred technical solution, the method further includes verifying the asymmetric knowledge distance, and the verification method includes:
[0031] Verifying the accuracy of the asymmetric knowledge distance based on the citation network and co-occurrence network;
[0032] Alternatively, the contribution is calculated based on the knowledge distance, and the accuracy of the asymmetric knowledge distance is verified according to the contribution value.
[0033] As a preferred technical solution, the method for verifying the accuracy of the asymmetric knowledge distance based on the citation network and the co-occurrence network is:
[0034] After removing the edges whose citation or co-occurrence frequency in the citation network and the co-occurrence network is lower than the preset frequency, edges in the citation network and the co-occurrence network are randomly selected according to the preset ratio as the training set, the validation set and the test set of the corresponding network prediction link connection;
[0035] Retraining the graph representation learning algorithm of the corresponding network using the training set and the validation set;
[0036] In the test set, the reference vector and co-occurrence vector are regenerated using the retrained graph representation learning algorithm, and a new fusion vector is generated, and the similarity value corresponding to the new fusion vector is calculated as the first link connection probability;
[0037] In the test set, the second link connection probability is calculated using traditional methods;
[0038] Calculate a first evaluation index based on the first link connection probability and the true value in the test set, and calculate a second evaluation index based on the second link connection probability and the true value in the test set;
[0039] The first evaluation index is compared with the second evaluation index to obtain the accuracy of the asymmetric knowledge distance.
[0040] As a preferred technical solution, the method for verifying the accuracy of the asymmetric knowledge distance according to the contribution value is:
[0041] Obtain the basic characteristics of patents. The basic characteristics and different types of knowledge distances constitute explanatory variables, and the patent influence is used as the explained variable. The basic characteristics include: patent application year, number of references, family size, number of applicants and number of inventors; the different types of knowledge distances are known values, including asymmetric knowledge distance, co-citation knowledge distance, inter-class knowledge distance and recombinant knowledge distance;
[0042] Predicting technological innovation performance using a baseline machine learning model based on the explanatory variables and the explained variables, and calculating the contribution value of each type of knowledge distance to the prediction of technological innovation performance;
[0043] The accuracy of each type of knowledge distance is verified based on the contribution value, that is, the greater the contribution value, the higher the accuracy.
[0044] According to a second aspect of the present invention, there is provided a technical knowledge distance measurement device based on graph representation learning, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above method when executing the program.
[0045] Compared with the prior art, the present invention has the following advantages:
[0046] 1) This application uses a graph representation learning algorithm to comprehensively consider the edge weights and direction information in the technical knowledge network, and fuses adjacent technical knowledge information into the target node vector in a biased form, accurately reflecting the asymmetry of technical knowledge;
[0047] 2) This application integrates semantic information, citation relationship and co-occurrence relationship into the same dimension, solving the problem that it is difficult to accurately capture knowledge distance by simply relying on traditional citation relationship and patent co-occurrence relationship, and can more accurately and comprehensively reflect knowledge distance;
[0048] 3) This application also provides two quantitative verification relationships to indirectly verify the accuracy of knowledge distance measurement. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Figure 1 A flowchart of a method for measuring technical knowledge distance based on graph representation learning according to the present invention;
[0050] Figure 2 A flowchart of obtaining a fusion vector according to the present invention;
[0051] Figure 3 This is a flow chart of verifying knowledge distance in Embodiment 2 of the present invention;
[0052] Figure 4 This is a flowchart for verifying knowledge distance in Example 3 of the present invention. DETAILED DESCRIPTION
[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.
[0054] Unless otherwise defined, the technical terms or scientific terms involved in this application should be understood by people with ordinary skills in the technical field to which this application belongs. The words "one", "a", "a", "the" and the like involved in this application do not indicate a quantitative limitation, and may represent the singular or plural. The terms "include", "comprise", "have" and any of their variations involved in this application are intended to cover non-exclusive inclusions; for example, a process, method, system, product or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units that are not listed, or may also include other steps or units inherent to these processes, methods, products or devices. The words "connect", "connected", "coupled" and the like involved in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. The "multiple" involved in this application refers to two or more. "And / or" describes the association relationship of associated objects, indicating that there may be three relationships, for example, "A and / or B" can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects before and after are in an "or" relationship. The terms "first", "second", "third", etc. involved in this application are only used to distinguish similar objects and do not represent a specific ordering of the objects.
[0055] Example 1
[0056] This embodiment provides a method for measuring technical knowledge distance based on graph representation learning to mine asymmetric relationships between technical knowledge and more accurately and comprehensively reflect knowledge distance through complete references, co-occurrences and semantic relationships. The process of this method is as follows: Figure 1 As shown, the detailed steps include:
[0057] S1. Obtain patent bibliometric data and extract technical knowledge elements based on the patent bibliometric data.
[0058] Among them, patent document measurement data includes patent classification number, application number, patent document, application date and patent name, etc.
[0059] S11. Structured patent bibliometric data and obtain patent classification numbers;
[0060] S12. Identify and extract the main classification number of the patent classification number, and extract technical knowledge elements based on the results of identification and extraction.
[0061] S2. Obtain the association relationship between technical knowledge elements, and construct a relationship network based on the association relationship, wherein the association relationship includes co-occurrence and citation. Therefore, the relationship network constructed based on the association relationship includes a co-occurrence network and a citation network.
[0062] S3. Extract the semantic information of each technical knowledge element.
[0063] S31, obtaining semantic interpretations of all sub-classification codes contained in the main classification code;
[0064] S32. Convert the semantic interpretation of each sub-classification number into a character string, and concatenate the character strings to obtain semantic information.
[0065] S4, based on the association network and the semantic information, obtain the reference vector, the relationship vector and the semantic vector, and fuse them to obtain the fusion vector. The process is as follows: Figure 2 As shown, the detailed steps are:
[0066] S41. Encode the nodes of the reference network and the co-occurrence network using the Node2Vec algorithm to obtain the co-occurrence vector and the reference vector.
[0067] S42. Based on semantic information, the semantic vector is extracted using BERT pre-trained large language model encoding technology.
[0068] S43, concatenating the reference vector, the co-occurrence vector and the semantic vector to obtain a fusion vector.
[0069] S431. Reduce the citation vector, co-occurrence vector and semantic vector to the same dimension.
[0070] S432, performing horizontal splicing of the reference vectors, co-occurrence vectors and semantic vectors after dimensionality reduction by horizontal expansion, wherein the reference vectors, co-occurrence vectors and semantic vectors after dimensionality reduction are all multi-dimensional vectors.
[0071] S5. Calculate the similarity of the fusion vectors, and calculate the asymmetric knowledge distance based on the similarity.
[0072] S51, calculating the similarity of the fusion vector using cosine similarity;
[0073] S52. Use Max-Min normalization to convert similarity into asymmetric knowledge distance, which is expressed as:
[0074]
[0075] Among them, KDi,j KS represents the asymmetric knowledge distance between technical knowledge element j and technical knowledge element i; i,j Represents the similarity between the fusion vector of knowledge technology element i and the fusion vector of j; Min i Indicates the minimum similarity between the fusion vector of technical knowledge element i and the fusion vectors of other technical knowledge elements; Max i Indicates the maximum similarity between the fusion vector of knowledge element i and the fusion vectors of other technical knowledge elements.
[0076] Example 2
[0077] This embodiment provides a method for verifying the accuracy of asymmetric knowledge distance based on citation network and co-occurrence network. The process is as follows: Figure 3 As shown, the detailed steps include:
[0078] A1. Because the network density in the citation network and the co-occurrence network may be high, it is impossible to extract the corresponding number of negative examples (i.e., there are no connected edges) in the verification set and the test set for evaluating the training performance of the Node2Vec algorithm. Therefore, in this embodiment, it is necessary to first remove the edges in the citation network and the co-occurrence network whose citation or co-occurrence frequency is lower than the preset frequency (usually set to 5), that is, the edges without link connections, so as to reduce the network density and ensure that there are enough negative examples in the citation network and the co-occurrence network.
[0079] A2. Randomly extract edges from the citation network and the co-occurrence network processed by step S51 at a ratio of 5%, 10% and 85% as the training set, validation set and test set of the corresponding network prediction link connection network.
[0080] A3. Using the training set and validation set, the grid search method is used to determine the optimal parameters of the Node2Vec algorithm, and the graph representation learning algorithm of the corresponding network is retrained.
[0081] A4. In the test set, use the retrained Node2Vec algorithm to regenerate the reference vector and co-occurrence vector, generate a new fusion vector, and calculate the similarity value corresponding to the new fusion vector as the first link connection probability;
[0082] A5. In the test set, the second link connection probability is calculated using traditional methods such as Spectral Clustering, LINE, DeepWalk, and GAE;
[0083] A6. Calculate a first evaluation index based on the first link connection probability and the true value in the test set, and calculate a second evaluation index based on the second link connection probability and the true value in the test set.
[0084] Specifically, the first evaluation index and the second evaluation index in step A6 are both AUC values and AP values, wherein the AUC index reflects whether the method provided in the above embodiment can significantly distinguish between existing links and non-existent links, and its formula is specifically as follows:
[0085]
[0086] N positive The graph represents the number of times the learning model correctly predicts the connection of new links, that is, the number of cases where the positive sample scores higher than the negative sample; N tie Represents the number of cases where the scores of positive samples and negative samples are equal; N represents the total number of all possible positive and negative sample pairs;
[0087] The AP index reflects the accuracy and recall of the Node2Vec algorithm in predicting links. It is usually expressed as the area under the PR curve (Precision-Recall Curve). The calculation formula is as follows:
[0088]
[0089] n is the number of links in the sorted candidate edge set; P(k) represents the precision of the links predicted as positive in the first k predictions; Δr(k) represents the incremental recall at the kth position, that is, the incremental impact of the positive samples predicted at this position on the overall number of positive samples.
[0090] A7. By comparing the two evaluation indicators calculated in step A6, the accuracy of the asymmetric knowledge distance can be verified.
[0091] Example 3
[0092] This embodiment provides a method for calculating contribution based on knowledge distance and verifying the accuracy of asymmetric knowledge distance according to the contribution value. The flow chart is as follows: Figure 4 As shown, the following steps are included in detail:
[0093] B1. Obtain the basic characteristics of the patent, including the year of patent application, number of patent references, size of patent family, number of patent applicants and number of patent inventors. The detailed information is shown in Table 1:
[0094] Table 1 Basic characteristics information of patents
[0095]
[0096]
[0097] B2. Obtain different types of knowledge distances, including asymmetric knowledge distance, co-citation knowledge distance, inter-class knowledge distance and reorganized knowledge distance. Detailed information is shown in Table 2:
[0098] Table 2 Different types of knowledge distance information
[0099]
[0100] B3. Take basic characteristics and different types of knowledge distance as explanatory variables, and patent influence as the explained variable. The detailed patent influence refers to the number of citations within five years of patent publication. The higher the citation frequency of a patent, the greater the influence of the patent and the higher the innovation performance.
[0101] B4. Use the baseline machine learning model to predict technological innovation performance based on explanatory variables and explained variables.
[0102] B41. This embodiment provides five baseline models for predicting technological innovation performance, including SVR, DecisionTree, RandomForest, XGB and AdaBoost. The model with the best performance among the five baseline models is selected as the prediction model. The detailed model information and hyperparameter description are shown in Table 3:
[0103] Table 3 Baseline model information table
[0104]
[0105] B42. Based on the explanatory variables and the explained variables, the best performing baseline model is used to predict the technological innovation performance. When making predictions, each type of knowledge distance is input into the best baseline model in batches and the changes in prediction performance are observed.
[0106] B5. Combine the four types of knowledge distances input into the baseline model, use the SHAP value to calculate the contribution value of each type of knowledge distance to predicting technological innovation performance, and verify the accuracy of the asymmetric knowledge distance based on the contribution value. Specifically, the larger the contribution value, the higher the accuracy.
[0107] Example 4
[0108] The above is an introduction to the method embodiment. The following is a further explanation of the scheme of the present invention through an equipment embodiment. Technical personnel in the relevant field can clearly understand that for the convenience and conciseness of description, the specific working process described can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.
[0109] The present embodiment provides a technical knowledge distance measurement device based on graph representation learning, which includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In RAM, various programs and data required for device operation can also be stored. CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.
[0110] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.
[0111] The processing unit performs the various methods and processes described above, such as methods S1 to S5, A1 to A7, and B1 to B5. For example, in some embodiments, methods S1 to S5, A1 to A7, and B1 to B5 may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into the RAM and executed by the CPU, one or more steps of methods S1 to S5, A1 to A7, and B1 to B5 described above may be performed. Alternatively, in other embodiments, the CPU may be configured to execute methods S1 to S5, A1 to A7, and B1 to B5 by any other appropriate means (e.g., by means of firmware).
[0112] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.
[0113] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.
[0114] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
[0115] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
Claims
1. A method for measuring technical knowledge distance based on graph representation learning, characterized in that: The method includes: Acquiring patent bibliometric data, and extracting technical knowledge elements based on the patent bibliometric data; Acquire the association relationship between the technical knowledge elements, and construct a relationship network based on the association relationship, wherein the association relationship includes co-occurrence and citation, and the relationship network includes a co-occurrence network and a citation network; Extract semantic information of each technical knowledge element; Based on the association network and the semantic information, a reference vector, a relationship vector and a semantic vector are obtained, and they are fused to obtain a fusion vector; The similarity of the fused vectors is calculated, and the asymmetric knowledge distance is calculated based on the similarity.
2. According to claim 1, a method for measuring technical knowledge distance based on graph representation learning is characterized in that: The method for extracting technical knowledge elements is: Structuring the patent bibliometric data and obtaining patent classification numbers; Identify and extract the main classification number of the patent classification number, and extract technical knowledge elements based on the results of identification and extraction.
3. According to claim 2, a method for measuring technical knowledge distance based on graph representation learning is characterized in that: The method for extracting semantic information of each technical knowledge element is: Obtaining semantic interpretations of all sub-classification codes contained in the main classification code; The semantic interpretation of each sub-classification number is converted into a character string, and the character strings are concatenated to obtain semantic information.
4. According to claim 1, a method for measuring technical knowledge distance based on graph representation learning is characterized in that: The method for obtaining the fusion vector is: Encoding the nodes of the citation network and the co-occurrence network using a graph representation learning algorithm to obtain a co-occurrence vector and a citation vector; Extracting a semantic vector based on the semantic information; The reference vector, co-occurrence vector and semantic vector are concatenated to obtain a fusion vector.
5. According to claim 4, a method for measuring technical knowledge distance based on graph representation learning is characterized in that: The splicing processing method is: The citation vector, co-occurrence vector and semantic vector are all reduced to the same dimension; The citation vectors, co-occurrence vectors and semantic vectors after dimensionality reduction are horizontally spliced by horizontal expansion, wherein the citation vectors, co-occurrence vectors and semantic vectors after dimensionality reduction are all multi-dimensional vectors.
6. The method for measuring technical knowledge distance based on graph representation learning according to claim 1, characterized in that: The method for calculating the asymmetric knowledge distance is: Use cosine similarity to calculate the similarity of the fused vectors; Max-Min normalization is used to convert the similarity into an asymmetric knowledge distance, which is expressed as: Among them, KD i,j KS represents the asymmetric knowledge distance between technical knowledge element j and technical knowledge element i; i,j Represents the similarity between the fusion vector of knowledge technology element i and the fusion vector of j; Min i Indicates the minimum similarity between the fusion vector of technical knowledge element i and the fusion vectors of other technical knowledge elements; Max i Indicates the maximum similarity between the fusion vector of knowledge element i and the fusion vectors of other technical knowledge elements.
7. The method for measuring technical knowledge distance based on graph representation learning according to claim 1, characterized in that: The method also includes verifying the asymmetric knowledge distance, and the verification method includes: Verifying the accuracy of the asymmetric knowledge distance based on the citation network and co-occurrence network; Alternatively, the contribution is calculated based on the knowledge distance, and the accuracy of the asymmetric knowledge distance is verified according to the contribution value.
8. The method for measuring technical knowledge distance based on graph representation learning according to claim 7, characterized in that: The method for verifying the accuracy of the asymmetric knowledge distance based on the citation network and the co-occurrence network is: After removing the edges whose citation or co-occurrence frequency in the citation network and the co-occurrence network is lower than the preset frequency, edges in the citation network and the co-occurrence network are randomly selected according to the preset ratio as the training set, the validation set and the test set of the corresponding network prediction link connection; Retraining the graph representation learning algorithm of the corresponding network using the training set and the validation set; In the test set, the reference vector and co-occurrence vector are regenerated using the retrained graph representation learning algorithm, and a new fusion vector is generated, and the similarity value corresponding to the new fusion vector is calculated as the first link connection probability; In the test set, the second link connection probability is calculated using traditional methods; Calculate a first evaluation index based on the first link connection probability and the true value in the test set, and calculate a second evaluation index based on the second link connection probability and the true value in the test set; The first evaluation index is compared with the second evaluation index to obtain the accuracy of the asymmetric knowledge distance.
9. The method for measuring technical knowledge distance based on graph representation learning according to claim 7, characterized in that: The method for verifying the accuracy of the asymmetric knowledge distance according to the contribution value is: Obtain the basic characteristics of patents. The basic characteristics and different types of knowledge distances constitute explanatory variables, and the patent influence is used as the explained variable. The basic characteristics include: patent application year, number of references, family size, number of applicants and number of inventors; the different types of knowledge distances are known values, including asymmetric knowledge distance, co-citation knowledge distance, inter-class knowledge distance and recombinant knowledge distance; Predicting technological innovation performance using a baseline machine learning model based on the explanatory variables and the explained variables, and calculating the contribution value of each type of knowledge distance to the prediction of technological innovation performance; The accuracy of each type of knowledge distance is verified based on the contribution value, that is, the greater the contribution value, the higher the accuracy.
10. A technical knowledge distance measurement device based on graph representation learning, comprising a memory and a processor, wherein a computer program is stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 9 is implemented.